mvNCCompile and mvNCCheck disagree

idata · ‎02-19-2018

HI everyone, i'm having some issues with NCS tools, my network compiles but if i try to check it i get this error:

mvNCCheck v02.00, Copyright @ Movidius Ltd 2016

USB: Transferring Data...
[Error 25] Myriad Error: "Softmax axis parameter = NULL".

Here is my prototxt hoping this helps:

input:"data"

input_shape{

dim: 1

dim: 63

dim: 13

}

layer {

type: "Convolution"

bottom: "data"

top: "conv1/dw"

convolution_param {

num_output: 1

bias_term: false

kernel_h: 15

kernel_w: 15

pad_w: 12

group: 1

#engine: CAFFE

stride: 1

weight_filler {

type: "constant"

value: 0

}

layer{

type: "Deconvolution"

bottom: "conv1/dw"

top: "Deconv1"

convolution_param {

num_output: 1

bias_term: false

kernel_size: 15

stride: 1

pad_w:13

pad_h:7

}

layer {

type: "Crop"

bottom: "conv1/dw"

bottom: "Deconv1"

top: "conv1/dw/cropped"

crop_param {

axis: -1

offset: 12

}

layer {
  name: "conv1/sep"
  type: "Convolution"
  bottom: "conv1/dw/cropped"
  top: "conv1/sep"
  convolution_param {
    num_output: 16
    bias_term: true
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv1/sep/bn"
  type: "BatchNorm"
  bottom: "conv1/sep"
  top: "conv1/sep/bn"
}
layer {
  name: "conv1/sep/bn/scale"
  type: "Scale"
  bottom: "conv1/sep/bn"
  top: "conv1/sep/bn"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}
layer {
  name: "relu1/sep"
  type: "ReLU"
  bottom: "conv1/sep/bn"
  top: "conv1/sep/rl"
}
layer {
  name: "conv2/dw"
  type: "Convolution"
  bottom: "conv1/sep/rl"
  top: "conv2/dw"
  convolution_param {
    num_output: 16
    bias_term: false
    pad_w: 7
    kernel_h: 10
    kernel_w: 10
    group: 16
    #engine: CAFFE
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer{
  name: "Deconv2"
  type: "Deconvolution"
  bottom: "conv2/dw"
  top: "Deconv2"
  convolution_param {
    num_output: 16
    bias_term: false
    kernel_size: 10
    stride: 1
    pad_w:8
    pad_h:4
  }
}
layer {
name: "conv2/dw/cropped"
type: "Crop"
bottom: "conv2/dw"
bottom: "Deconv2"
top: "conv2/dw/cropped"
crop_param {
axis: -1
offset: 7
}
}

layer {
  name: "conv2/sep"
  type: "Convolution"
  bottom: "conv2/dw/cropped"
  top: "conv2/sep"
  convolution_param {
    num_output: 64
    bias_term: true
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "conv2/sep/bn"
  type: "BatchNorm"
  bottom: "conv2/sep"
  top: "conv2/sep"
}
layer {
  name: "conv2/sep/bn/scale"
  type: "Scale"
  bottom: "conv2/sep"
  top: "conv2/sep"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}

layer {
  name: "relu2/sep"
  type: "ReLU"
  bottom: "conv2/sep"
  top: "conv2/sep"
}
layer {
  name: "conv3/dw"
  type: "Convolution"
  bottom: "conv2/sep"
  top: "conv3/dw"
  convolution_param {
    num_output: 64
    bias_term: false
    pad_w: 2 
    kernel_h: 5
    kernel_w: 5
    group: 64
    #engine: CAFFE
    stride: 1
    weight_filler {
      type: "constant"
      value: 0
    }
  }
}
layer{
  name: "Deconv3"
  type: "Deconvolution"
  bottom: "conv3/dw"
  top: "Deconv3"
  convolution_param {
    num_output: 64
    bias_term: false
    kernel_size: 5
    stride: 1
    pad_w:3
    pad_h:2
  }
}
layer {
name: "conv3/dw/cropped"
type: "Crop"
bottom: "conv3/dw"
bottom: "Deconv3"
top: "conv3/dw/cropped"
crop_param {
axis: -1
offset: 2
}
}
layer {
  name: "conv3/sep"
  type: "Convolution"
  bottom: "conv3/dw/cropped"
  top: "conv3/sep"
  convolution_param {
    num_output: 128
    bias_term: true
    pad: 0
    kernel_size: 1
    stride: 1
    weight_filler {
      type: "msra"
    }
  }
}
layer {
  name: "conv3/sep/bn"
  type: "BatchNorm"
  bottom: "conv3/sep"
  top: "conv3/sep"
}
layer {
  name: "conv3/sep/bn/scale"
  type: "Scale"
  bottom: "conv3/sep"
  top: "conv3/sep"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}

layer {
  name: "relu3/sep"
  type: "ReLU"
  bottom: "conv3/sep"
  top: "conv3/sep"
}
layer {
  name: "avg_pool"
  type: "Pooling"
  bottom: "conv3/sep"
  top: "avg_pool"
  pooling_param {
    pool: AVE
    global_pooling: true
  }
}
layer {
  name: "fc"
  type: "InnerProduct"
  bottom: "avg_pool"
  top: "fc"
  inner_product_param {
    num_output: 12
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "fc/bn"
  type: "BatchNorm"
  bottom: "fc"
  top: "fc"
}
layer {
  name: "fc/bn/scale"
  type: "Scale"
  bottom: "fc"
  top: "fc"
  scale_param {
    filler {
      value: 1
    }
    bias_term: true
    bias_filler {
      value: 0
    }
  }
}

layer {
  name: "output"
  type: "Softmax"
  bottom: "fc"
  top: "output"
}

idata · ‎02-21-2018

@Ryose I'm not sure if this is the cause of your issue, but while examining your prototxt file I noticed that in layer conv2/dw you have a depthwise convolution of 10x10 and currently we only have support for depth-wise convolutions that are 3x3 (i.e. SSD Mobilenet Caffe).

idata · ‎02-22-2018

@Tome_at_Intel So if i specify the group attribute of a caffe layer, i can only use 3x3Conv, is that right?

idata · ‎02-22-2018

@Ryose Yes, to be more specific, if the group parameter is set to more than 1, you can only use 3x3 convolutions. Please let me know if you run into any other issues.